Futility + WAC

Archive of the old Parsimony forum. Some messages couldn't be restored. Limitations: Search for authors does not work, Parsimony specific formats do not work, threaded view does not work properly. Posting is disabled.

Futility + WAC

Postby Stefan Knappe » 07 Nov 2001, 09:23

Geschrieben von: / Posted by: Stefan Knappe at 07 November 2001 09:23:56:
Hi,
last night my program tried to solve the WAC test suite. The first time it worked with and the second time without futility prunning.
I was suprised to see, the version without prunning solved more positions than the version with prunning.
Did anybody see the same behavior? Do I have to look for a bug in my code?
Cheers
Stefan
Stefan Knappe
 

Re: Futility + WAC

Postby Thomas Mayer » 07 Nov 2001, 09:30

Geschrieben von: / Posted by: Thomas Mayer at 07 November 2001 09:30:18:
Als Antwort auf: / As an answer to: Futility + WAC geschrieben von: / posted by: Stefan Knappe at 07 November 2001 09:23:56:
Hi Stefan,
last night my program tried to solve the WAC test suite. The first time it
worked with and the second time without futility prunning.
I was suprised to see, the version without prunning solved more positions
than the version with prunning.
Did anybody see the same behavior? Do I have to look for a bug in my code?
thats quite usual - futility prunning does weaken your tactical ability in some points especially when there are initial sacrifices... (Just think about the algorithm - it's easy to understand) - so you will find some solutions one ply later, with limited razoring and extended futility pruning up to three plys later or more if some null move issues also comes in mind - prunning is dangerous... :)
Greets, Thomas
Thomas Mayer
 

Re: Futility + WAC

Postby Stefan Knappe » 07 Nov 2001, 10:13

Geschrieben von: / Posted by: Stefan Knappe at 07 November 2001 10:13:10:
Als Antwort auf: / As an answer to: Re: Futility + WAC geschrieben von: / posted by: Thomas Mayer at 07 November 2001 09:30:18:
thats quite usual - futility prunning does weaken your tactical ability in some points especially when there are initial sacrifices... (Just think about the algorithm - it's easy to understand) - so you will find some solutions one ply later, with limited razoring and extended futility pruning up to three plys later or more if some null move issues also comes in mind - prunning is dangerous... :)
Hi Thomas,
yes, I know. I believe I understood the algorithm well :-) But after reading the book by A. Heinz I thought this kind of pruning would improve the tactical abilities because of the additional search depth.
Do you disable the pruning during Quark is solving a test suite?
Regards
Stefan
Stefan Knappe
 

Re: Futility + WAC

Postby Gian-Carlo Pascutto » 07 Nov 2001, 10:28

Geschrieben von: / Posted by: Gian-Carlo Pascutto at 07 November 2001 10:28:16:
Als Antwort auf: / As an answer to: Re: Futility + WAC geschrieben von: / posted by: Stefan Knappe at 07 November 2001 10:13:10:
thats quite usual - futility prunning does weaken your tactical ability in some points especially when there are initial sacrifices... (Just think about the algorithm - it's easy to understand) - so you will find some solutions one ply later, with limited razoring and extended futility pruning up to three plys later or more if some null move issues also comes in mind - prunning is dangerous... :)
Hi Thomas,
yes, I know. I believe I understood the algorithm well :-) But after reading the book by A. Heinz I thought this kind of pruning would improve the tactical abilities because of the additional search depth.
Do you disable the pruning during Quark is solving a test suite?
Regards
Stefan
This differs a whole lot from program to program.
For some it helps, for others it hurts.
For some it doesn't make a difference.
--
GCP
Gian-Carlo Pascutto
 

Re: Futility + WAC

Postby Thomas Mayer » 07 Nov 2001, 10:43

Geschrieben von: / Posted by: Thomas Mayer at 07 November 2001 10:43:01:
Als Antwort auf: / As an answer to: Re: Futility + WAC geschrieben von: / posted by: Stefan Knappe at 07 November 2001 10:13:10:
Hi Stefan,
thats quite usual - futility prunning does weaken your tactical ability in
some points especially when there are initial sacrifices... (Just think
about the algorithm - it's easy to understand) - so you will find some
solutions one ply later, with limited razoring and extended futility pruning
up to three plys later or more if some null move issues also comes in mind -
prunning is dangerous... :)
yes, I know. I believe I understood the algorithm well :-) But after reading
the book by A. Heinz I thought this kind of pruning would improve the
tactical abilities because of the additional search depth.
Do you disable the pruning during Quark is solving a test suite?
of course not... Else the result would be not very helpful... :) I have different prunning then futility-prunning... But Gian-Carlo is correct, for some it helps, for some it hurts, depends on to many things - try WAC with different times, maybe it will hurt you with 5 secs but help with 10 secs and look in Ernst A. Heinz book, it is also commented, that it does solve not the same amount of positions in WAC. So he knows about the problem but thinks that for his program it is overall a win... That's why I always say if you hear a new idea: "try it, maybe it works, maybe not - but after that you can be sure that you have tried..." :)
In newer versions I might try Heinz's ideas, at the moment I try to catch up a bug in search somewhere which is not fully discovered by me at the moment... it's very strange and deep in the tree which does not make it easy to debug... but I have already two positions where it occurs, so maybe I can find it...
Greets, Thomas
Thomas Mayer
 

Re: Futility + WAC

Postby Stefan Knappe » 07 Nov 2001, 11:06

Geschrieben von: / Posted by: Stefan Knappe at 07 November 2001 11:06:16:
Als Antwort auf: / As an answer to: Re: Futility + WAC geschrieben von: / posted by: Thomas Mayer at 07 November 2001 10:43:01:
I will try it again with more time!
In newer versions I might try Heinz's ideas, at the moment I try to catch up a bug in search somewhere which is not fully discovered by me at the moment... it's very strange and deep in the tree which does not make it easy to debug... but I have already two positions where it occurs, so maybe I can find it...
You shold not spend your time with searching this bug! Maybe it is only one unimportant bug! I would like to help you to find it after our next game in Paderborn ;-)
Thanks,
Stefan
Stefan Knappe
 

Re: Futility + WAC

Postby U.Tuerke » 07 Nov 2001, 11:43

Geschrieben von: / Posted by: U.Tuerke at 07 November 2001 11:43:34:
Als Antwort auf: / As an answer to: Re: Futility + WAC geschrieben von: / posted by: Thomas Mayer at 07 November 2001 09:30:18:
Hi Stefan,
last night my program tried to solve the WAC test suite. The first time it
worked with and the second time without futility prunning.
I was suprised to see, the version without prunning solved more positions
than the version with prunning.
Did anybody see the same behavior? Do I have to look for a bug in my code?
thats quite usual - futility prunning does weaken your tactical ability in some points especially when there are initial sacrifices... (Just think about the algorithm - it's easy to understand) - so you will find some solutions one ply later, with limited razoring and extended futility pruning up to three plys later or more if some null move issues also comes in mind - prunning is dangerous... :)
Greets, Thomas
I agree. IMHO, it's very import to identify positions where better to switch off the forward pruning (hanging pieces, in check, king attack running, ...) algos.
In my experience, these forward pruning attempts are a very double-edged thing. It requires a lot of testing and tuning to make them reasonably. If you don't make these reasonably, you'd better skip it. It's very easy to spoil your program's play this way. Large search depths do impress the user, but it's not guaranteed that they really qualitatively improve the search.
I'm not sure if I do these cuts okay and have often considered removing them.
But I still try these cuts, and they save some nodes.
Regards,
Uli
U.Tuerke
 

Re: Futility + WAC

Postby Thomas Mayer » 07 Nov 2001, 13:01

Geschrieben von: / Posted by: Thomas Mayer at 07 November 2001 13:01:37:
Als Antwort auf: / As an answer to: Re: Futility + WAC geschrieben von: / posted by: U.Tuerke at 07 November 2001 11:43:34:
Hi Uli,
thats quite usual - futility prunning does weaken your tactical ability in
some points especially when there are initial sacrifices... (Just think
about the algorithm - it's easy to understand) - so you will find some
solutions one ply later, with limited razoring and extended futility pruning
up to three plys later or more if some null move issues also comes in mind -
prunning is dangerous... :)
I agree. IMHO, it's very import to identify positions where better to switch
off the forward pruning (hanging pieces, in check, king attack running, ...)
algos.
In my experience, these forward pruning attempts are a very double-edged
thing. It requires a lot of testing and tuning to make them reasonably. If
you don't make these reasonably, you'd better skip it. It's very easy to
spoil your program's play this way. Large search depths do impress the user,
but it's not guaranteed that they really qualitatively improve the search.
I'm not sure if I do these cuts okay and have often considered removing them.
But I still try these cuts, and they save some nodes.
same to me - I do not use futility prunning, my prunning system is a little bit inspired by that what Gnu is doing with some different rules when to use it and when better to turn it off... When I ever will get the attacktables ready while deepening and not only in evaluation I will try some more prunnings - I think prunning is the best way to get big efforts in playing strength if you get them reasonable... but to play in tourneys WITHOUT fully test them might be no good idea... :) In Berlin we discuss much about Tiger and its incredible depth, we all fully agree that those depth can only be reached by forward prunning - the question is: how is he doing them in such a reasonable way... hehe, still hoping to get once a branching factor below 2... :))) (Vincent would say Diep is already below 1... :)
Greets, Thomas
Thomas Mayer
 

Re: Futility + WAC

Postby Jose Carlos » 07 Nov 2001, 13:27

Geschrieben von: / Posted by: Jose Carlos at 07 November 2001 13:27:06:
Als Antwort auf: / As an answer to: Re: Futility + WAC geschrieben von: / posted by: Thomas Mayer at 07 November 2001 09:30:18:
Hi Stefan,
last night my program tried to solve the WAC test suite. The first time it
worked with and the second time without futility prunning.
I was suprised to see, the version without prunning solved more positions
than the version with prunning.
Did anybody see the same behavior? Do I have to look for a bug in my code?
thats quite usual - futility prunning does weaken your tactical ability in some points especially when there are initial sacrifices... (Just think about the algorithm - it's easy to understand) - so you will find some solutions one ply later, with limited razoring and extended futility pruning up to three plys later or more if some null move issues also comes in mind - prunning is dangerous... :)
Greets, Thomas
I must add that testsuites are very different from chess games. In tests, you have to find a best move (or avoid a bad one) in every position. In a game, you don't have a best move except for a few times. Most positions require only sound moves, not making tactical mistakes and play according to some strategic rules.
You can make your program overall stronger even if it's slower solving test positions. Tests help finding bugs or big mistakes in search or eval, but can't really measure how strong the program will play in real games.
José C.
Jose Carlos
 

Re: Futility + WAC

Postby Miguel A. Ballicora » 07 Nov 2001, 15:26

Geschrieben von: / Posted by: Miguel A. Ballicora at 07 November 2001 15:26:02:
Als Antwort auf: / As an answer to: Re: Futility + WAC geschrieben von: / posted by: Thomas Mayer at 07 November 2001 09:30:18:
Hi Stefan,
last night my program tried to solve the WAC test suite. The first time it
worked with and the second time without futility prunning.
I was suprised to see, the version without prunning solved more positions
than the version with prunning.
Did anybody see the same behavior? Do I have to look for a bug in my code?
thats quite usual - futility prunning does weaken your tactical ability in some points especially when there are initial sacrifices... (Just think about the algorithm - it's easy to understand) - so you will find some solutions one ply later, with limited razoring and extended futility pruning up to three plys later or more if some null move issues also comes in mind - prunning is dangerous... :)
Greets, Thomas
As an example, futility pruning has a bad behavior when combined with nullmove and positions where you threat, after a sacrifice, a checkmate that is unstoppable. That checkmate (that is not a capture!) can pruned in the nullmove search and the solution is delayed a lot. This kind of situations are common in tactical attacks to the king (as in WAC).
Regards,
Miguel
Miguel A. Ballicora
 


Return to Archive (Old Parsimony Forum)

Who is online

Users browsing this forum: No registered users and 21 guests